I understand why most people are saying that AlphaGo is not brute force,
because it appears to be highly selective. But MCTS is a full width search.
Read the AlphaGo papers, as one of the other respondents (rather sarcastically)
suggested: AlphaGo will eventually search every move at every node.
MCTS has the appearance of a selective search because time control terminates
search while the tree is still ragged. In fact, it will search every
continuation an infinite number of times.
In order to have high performance, an MCTS implementation needs to search best
moves as early as possible in each node. It is in this respect that AlphaGo
truly excels. (AlphaGo also excels at whole board evaluation, but that is a
From: Steven Clark [mailto:steven.p.cl...@gmail.com]
Sent: Sunday, August 6, 2017 1:14 PM
To: Brian Sheppard <sheppar...@aol.com>; computer-go
Subject: Re: [Computer-go] Alphago and solving Go
Why do you say AlphaGo is brute-force? Brute force is defined as: "In computer
science, brute-force search or exhaustive search, also known as generate and
test, is a very general problem-solving technique that consists of
systematically enumerating all possible candidates for the solution and
checking whether each candidate satisfies the problem's statement."
The whole point of the policy network is to avoid brute-force search, by
reducing the branching factor...
On Sun, Aug 6, 2017 at 10:42 AM, Brian Sheppard via Computer-go
<firstname.lastname@example.org <mailto:email@example.com> > wrote:
Yes, AlphaGo is brute force.
No it is impossible to solve Go.
Perfect play looks a lot like AlphaGo in that you would not be able to tell the
difference. But I think that AlphaGo still has 0% win rate against perfect play.
My own best guess is that top humans make about 12 errors per game. This is
estimated based on the win rate of top pros in head-to-head games. The
calculation starts by assuming that Go is a win at 6.5 komi for either Black
(more likely) or White, so a perfect player would win 100% for Black. Actual
championship caliber players win 51% to 52% for Black. In 9-dan play overall, I
think the rate is 53% to 54% for Black. Then you can estimate how many errors
each player has to make to bring about such a result. E.g., If players made
only one error on average, then Black would win the vast majority of games, so
they must make more errors. I came up with 12 errors per game, but you can
reasonably get other numbers based on your model.
From: Computer-go [mailto:computer-go-boun...@computer-go.org
<mailto:computer-go-boun...@computer-go.org> ] On Behalf Of Cai Gengyang
Sent: Sunday, August 6, 2017 9:49 AM
To: firstname.lastname@example.org <mailto:email@example.com>
Subject: [Computer-go] Alphago and solving Go
Is Alphago brute force search?
Is it possible to solve Go for 19x19 ?
And what does perfect play in Go look like?
How far are current top pros from perfect play?
Computer-go mailing list
Computer-go mailing list